International Journal of Control Sliding-mode control of ...emilia/IJC10.pdf · content ... A...
Transcript of International Journal of Control Sliding-mode control of ...emilia/IJC10.pdf · content ... A...
PLEASE SCROLL DOWN FOR ARTICLE
This article was downloaded by: [Steklyar, Inna]On: 15 November 2010Access details: Access Details: [subscription number 920179271]Publisher Taylor & FrancisInforma Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK
International Journal of ControlPublication details, including instructions for authors and subscription information:http://www.informaworld.com/smpp/title~content=t713393989
Sliding-mode control of uncertain systems in the presence of unmatcheddisturbances with applicationsX. Hana; E. Fridmanb; S. K. Spurgeona
a Instrumentation, Control and Embedded Systems Research Group, School of Engineering and DigitalArts, Kent University, Canterbury, Kent, UK b School of Electrical Engineering, Tel Aviv University,Tel Aviv, Israel
First published on: 15 November 2010
To cite this Article Han, X. , Fridman, E. and Spurgeon, S. K.(2010) 'Sliding-mode control of uncertain systems in thepresence of unmatched disturbances with applications', International Journal of Control,, First published on: 15November 2010 (iFirst)To link to this Article: DOI: 10.1080/00207179.2010.527375URL: http://dx.doi.org/10.1080/00207179.2010.527375
Full terms and conditions of use: http://www.informaworld.com/terms-and-conditions-of-access.pdf
This article may be used for research, teaching and private study purposes. Any substantial orsystematic reproduction, re-distribution, re-selling, loan or sub-licensing, systematic supply ordistribution in any form to anyone is expressly forbidden.
The publisher does not give any warranty express or implied or make any representation that the contentswill be complete or accurate or up to date. The accuracy of any instructions, formulae and drug dosesshould be independently verified with primary sources. The publisher shall not be liable for any loss,actions, claims, proceedings, demand or costs or damages whatsoever or howsoever caused arising directlyor indirectly in connection with or arising out of the use of this material.
International Journal of Control2010, 1–14, iFirst
Sliding-mode control of uncertain systems in the presence of unmatched
disturbances with applications
X. Hana, E. Fridmanb and S.K. Spurgeona*
aInstrumentation, Control and Embedded Systems Research Group, School of Engineeringand Digital Arts, Kent University, Canterbury, Kent, UK; bSchool of Electrical Engineering,
Tel Aviv University, Tel Aviv, Israel
(Received 2 June 2010; final version received 23 September 2010)
This article considers the development of constructive sliding-mode control strategies based on measured outputinformation only for linear, time-delay systems with bounded disturbances that are not necessarily matched.The novel feature of the method is that linear matrix inequalities are derived to compute solutions to both theexistence problem and the finite time reachability problem that minimise the ultimate bound of the reduced-ordersliding-mode dynamics in the presence of state time-varying delay and unmatched disturbances. Themethodology provides guarantees on the level of closed-loop performance that will be achieved by uncertainsystems which experience delay. The methodology is also shown to facilitate sliding-mode controller design forsystems with polytopic uncertainties, where the uncertainty may appear in all blocks of the system matrices.A time-delay model with polytopic uncertainties from the literature provides a tutorial example of the proposedmethod. A case study involving the practical application of the design methodology in the area of autonomousvehicle control is also presented.
Keywords: sliding-mode control; output feedback; state delay; LMIs
1. Introduction
The control of time-delay systems is known to be ofpractical significance. Problems largely fall into twocategories. The first category arises because of the needto model systems more accurately given increasingperformance expectations. Many processes, such asmanufacturing processes, include such after effectphenomena in their dynamics and time delay is alsoproduced via the actuators, sensors and networksinvolved in the practical implementation of feedbackcontrol strategies. The second class of problems ariseswhen time delays are used as a modelling tool tosimplify some infinite-dimensional systems. Thisapproach is used for constructing models of distributedsystems modelled by partial differential equationswhere a set of finite-dimensional state variables withappropriate time-delay characteristics can be used torepresent heat exchange processes, for example.
The application of sliding-mode control (SMC) tothe problem of systems with time delay is a far fromtrivial problem generically, involving the combinationof delay phenomenon with relay actuators which hasthe potential to induce oscillations around the slidingsurface during the sliding-mode. There are a numberof papers which have considered the problem.
The development of sliding-mode controllers for
operation in the presence of single or multiple,
constant or time-varying state delays was solved by
Gouaisbaut, Dambrine, and Richard (2002). This uses
the equivalent control method where the exact knowl-
edge of block matrices must be known and the full-
state measurable. The work was further extended to
include polytopic uncertainties (Gouaisbaut, Blanco,
and Richard 2004), however full-state feedback was
required to yield finite reachability design and the
upper bound on the states spanning the input space is
assumed to be known for solution of the reachability
problem. The problem was also considered by Li and
DeCarlo (2003) where a class of uncertain time-delay
systems with multiple fixed delays in the system states
is considered. The method of equivalent control is used
and measurement of the state at the current time is
assumed. It is also assumed that the delayed state is
bounded with a bound dependent on the current state
which is restrictive. Even though an adaptive law was
used to estimate the bound of the uncertainty in the
reaching phase, a switching gain which is again
assumed to be large enough rather than calculated
explicitly is thought to guarantee the reachability
design. The work in Jafarov (2005) considers
*Corresponding author. Email: [email protected]
ISSN 0020–7179 print/ISSN 1366–5820 online
� 2010 Taylor & Francis
DOI: 10.1080/00207179.2010.527375
http://www.informaworld.com
Downloaded By: [Steklyar, Inna] At: 07:06 15 November 2010
sliding-mode control of an uncertain system in thepresence of fixed state delay, but again full-statefeedback is assumed.
The assumption of full-state feedback is a limitingone in practice as it may be prohibitively expensive, andindeed, sometimes impossible, to measure all the statevariables. One approach to solve this problem is toimplement the controller with an observer, wherethe observer provides state estimates for use bythe controller. However, the implementation of thecontroller–observer is more involved and the theoreticalframeworks to ensure stability across the range ofpractical operation of the plant may be challenging.A more straightforward approach is to use only thesubset of state information that is available, i.e. themeasured output, within the control design paradigm.
There are typically two facets to the design of astatic output feedback sliding-mode control. One is theexistence problem, i.e. the design of a switching surfacein the output vector space which is usually of lowerorder than the state vector space. Consider first theswitching surface design problem for uncertain sys-tems, where time-delay effects are not involved. Twodifferent methods of designing sliding surfaces usingeigenvalue assignment and eigenvector techniques wereproposed in Zak and Hui (1993) and El-Khazali andDecarlo (1995). A canonical form via which the staticoutput feedback sliding-mode control design problemis converted to a static output feedback stabilisationproblem for a particular subsystem triple was providedin Edwards and Spurgeon (1995). However, the solu-tion to the general static output feedback problem,even for linear time-invariant systems, is still open.Iterative linear matrix inequality (LMI) approacheshave been exploited to solve the static output feedbackproblem using a bilinear matrix inequality formula-tion, see Cao, Lam, and Sun (1998), Choi (2002) andHuang and Nguang (2006). In Edwards (2004), wherethe regular form was not used for synthesis of thecontrol law, LMIs were derived for switching functiondesign whilst minimising the cost function associatedwith the control. Sufficient conditions for static outputfeedback controller design using LMIs have also beensought. Although only sufficient, the solutions havethe advantage of being linear and, hence, easilytractable using standard optimisation techniques, seeCrusius and Trofino (1999) and Shaked (2003). Thesecond facet in the design of a sliding-mode outputfeedback controller is the control, or reachability,synthesis problem whereby a control is determined toensure the sliding surface is attractive. It is non-trivialto synthesise a control law only using the outputvector, even for the situation where time-delay effectsare not considered, since the derivative of the slidingsurface is always related to the unmeasured states and
this derivative appears in the reachability condition. Aswell as within the existence problem, LMI methodshave also been considered within the context ofdeveloping a sliding-mode control strategy whichsolves the reachability problem for a given slidingsurface. For example, LMI methods which yieldreachability conditions for designing static sliding-mode output feedback controllers are presented inEdwards, Akoachere, and Spurgeon (2001).
In the context of output feedback sliding-modecontrol for time-delay systems, the existence andreachability problems for systems in the presence ofmatched uncertainty are considered in Han, Fridman,Spurgeon, and Edwards (2009). The delay is assumedto be time-varying and bounded where the upperbound is known. In line with the development ofoutput feedback controllers in the non-delayed case,LMIs are used to select all the parameters of theclosed-loop sliding-mode controller. However, noexplicit calculation of the switching gain in thenonlinear part of the control was given, it was onlyassumed to be large enough to induce the sliding mode.While asymptotic stability in the presence of matcheddisturbances can be achieved by sliding-mode control,unmatched disturbances usually lead to only boundedstability. For example, in Fernando and Fridman(2006) robustness properties of integral sliding-modecontrollers are studied where the Euclidean norm ofthe unmatched perturbation is minimised by selecting aprojection matrix.
The contribution of the proposed work is that allthe design parameters, including the switching gain, arederived from LMIs in spite of the presence of statedelay and unmatched disturbances. The method is ableto deal with polytopic type uncertainties in all blocksof the system matrices. No additional assumption ismade on the bound of the uncertain states in thereachability design, as required by other work. Somepreliminary results of this article are presented in Hanet al. (2009). Central to the work presented is thedescriptor approach (Fridman 2001), which is appliedto derive LMIs for the solution of the sliding-modeoutput feedback control problem in the presence ofboth matched and unmatched bounded disturbancesand time-varying state delays. It is demonstrated thatthe state trajectories of the system converge towards aball with a prespecified exponential convergence rate.In Section 2, the problem formulation is described andan appropriate general framework to accomplish theoutput feedback sliding-mode control design is formu-lated in Section 3. A constructive solution to theexistence problem is presented in Section 4 and Section5 shows the formulation of the reachability problemwhich will ensure that the sliding-mode is reached. Aproblem from the literature is used to provide a
2 X. Han et al.
Downloaded By: [Steklyar, Inna] At: 07:06 15 November 2010
tutorial example of how the paradigm can be used tosolve both the existence and reachability problems forpractical design. A case study relating to the control ofan autonomous vehicle is used to further illustrate thedesign process in Section 6.
Notation: Rn denotes the n-dimensional Euclideanspace with vector norm k � k or the induced matrixnorm, Rn�m is the set of all n�m real matrices. P4 0,for P2Rn�n, denotes that P is symmetric and positivedefinite whereas * means the symmetric entries ofan LMI.
2. Problem formulation
Consider an uncertain dynamical system of the form
_xðtÞ ¼ AxðtÞ þ Adxðt� �ðtÞÞ þ BuðtÞ þ B1wðtÞ,
yðtÞ ¼ CxðtÞ,
xðt0 � �ðtÞÞ ¼ �ð�ðtÞÞ for �ðtÞ 2 ½0, h�,
ð1Þ
where x2Rn, u2Rm, w2Rk and y2Rp withm5 p5 n, � is absolutely continuous with squareintegrable _�, h is an upper bound on the time-delayfunction (0� �(t)� h 8t� 0). The time-varying delaymay be either slowly varying (i.e. a differentiabledelay with _�ðtÞ � d5 1) or fast varying (piecewise-continuous delay). Assume that the nominal linearsystem (A,Ad,B,B1,C ) is known and that the input andoutput matrices B and C are both of full rank. Thedisturbance is assumed to be bounded wherebykw(t)k�D with a known upper bound D4 0. A controlstrategy will be sought which induces an ideal slidingmotion with desirable performance characteristics onthe surface
S ¼ fx 2 Rn : sðtÞ ¼ FCxðtÞ ¼ 0g ð2Þ
for some selected matrix F2Rm�p so that the motion,when restricted to S, is stable.
3. A general framework for design
The first problem considered is how to choose F in(2) so that the associated sliding motion is stable. Acontrol law will then be determined to guarantee theexistence of a sliding motion. A convenient systemrepresentation closely allied to the usual regularform used for sliding-mode control design isemployed. It can be shown that if rank(CB)¼mand system triple (A,B,C ) are minimum phase, thereexists a coordinate system xr¼Trx, xr¼ [x1 x2]
T, in
which the system (A,Ad,B,B1,C ) has the trans-
formed structure
Ar ¼A11 A12
A21 A22
" #, Adr ¼
Ad11 Ad12
Ad21 Ad22
" #,
Br ¼0
B2
" #, B1r ¼
B11
B12
" #, Cr ¼ 0 T
� �, ð3Þ
where B22Rm�m is nonsingular and T2Rp�p is
orthogonal (Edwards and Spurgeon 1995).
Furthermore, A11, Ad112R(n�m)� (n�m) and the remain-
ing sub-blocks are partitioned accordingly. Let
F1 F2
� �¼ FT, ð4Þ
where F12Rm�( p�m) and F22R
m�m. As a result
FCr ¼ F1C1 F2
� �, ð5Þ
where
C1 ¼ 0ð p�mÞ�ðn�pÞ Ið p�mÞ� �
: ð6Þ
It is straightforward to see that FCrBr¼F2B2 and the
square matrix F2 is non-singular. By assumption, the
system contains both matched and unmatched uncer-
tainties and therefore the sliding motion is independent
of the matched uncertainty but dependent on the
unmatched uncertainty. In terms of the coordinate
framework defined above, the reduced-order sliding-
mode dynamics are governed by the following reduced-
order system
_x1ðtÞ ¼ ðA11 � A12KC1Þx1ðtÞ þ ðAd11 � Ad12KC1Þ
� x1ðt� �ðtÞÞ þ B11wðtÞ: ð7Þ
The response of this system must therefore be ulti-
mately bounded, where K ¼ F�12 F1, and the problem
of hyperplane design is equivalent to a static output
feedback problem for the system (A11,Ad11,A12,Ad12,
C1), where (A11þAd11,A12þAd12) is assumed control-
lable and (A11þAd11,C1) observable. Note that
the presence of the unmatched uncertainty means
that, in general, asymptotic stability cannot be
attained by the system (7). This is formalised in terms
of the existence problem, which must be solved to
determine the switching surface, in the following
section.
4. Existence problem
It will be shown that the system (3) is exponentially
attracted to a bounded region in Rn if the reduced-
order system (7) is exponentially attracted to a bounded
domain in Rn�m. Consider the Lyapunov–Krasovskii
International Journal of Control 3
Downloaded By: [Steklyar, Inna] At: 07:06 15 November 2010
functional below for the exponential stability analysis
of (7)
VðtÞ ¼ xT1 ðtÞPx1ðtÞ þ
Z t
t�h
e�ðs�tÞxT1 ðsÞEx1ðsÞds
þ
Z t
t��ðtÞ
e�ðs�tÞxT1 ðsÞSx1ðsÞds
þ h
Z 0
�h
Z t
tþ�
e�ðs�tÞ _xT1 ðsÞR _x1ðsÞds d� ð8Þ
with (n�m)� (n�m)-matrices P4 0 and E� 0, S� 0,
R� 0. To prove exponential stability of the system (7)
using (8), we will use the following lemma.
Lemma 4.1 (Fridman and Dambrine 2009): Let V :
[0,1)!Rþ be an absolutely continuous function. If
there exist �4 0 and b4 0 such that the derivative of V
satisfies almost everywhere the inequality
d
dtVðtÞ þ �VðtÞ � bkwðtÞk2 � 0
then it follows that for all kw(t)k�D
VðtÞ � e��ðt�t0ÞVðt0Þ þb
�D2, t � t0:
Differentiating V(t) from (8) yields
M � 2xT1 ðtÞP _x1ðtÞ þ h2 _xT1 ðtÞR _x1ðtÞ
� he��hZ t
t�h
_xT1 ðsÞR _x1ðsÞdsþ xT1 ðtÞðEþ S Þx1ðtÞ
� xT1 ðt� hÞEx1ðt� hÞe��h þ �xT1 ðtÞPx1ðtÞ
� ð1� _�ðtÞÞxT1 ðt� �ðtÞÞSx1ðt� �ðtÞÞe���ðtÞ
� bwTðtÞwðtÞ: ð9Þ
Further using the identity
� h
Z t
t�h
_xT1 ðsÞR _x1ðsÞds
¼�h
Z t��ðtÞ
t�h
_xT1 ðsÞR _x1ðsÞds� h
Z t
t��ðtÞ
_xT1 ðsÞR _x1ðsÞds
ð10Þ
and applying Jensen’s inequalityZ t
t��ðtÞ
_xT1 ðsÞR _x1ðsÞds �1
h
Z t
t��ðtÞ
_xT1 ðsÞdsR
Z t
t��ðtÞ
_x1ðsÞds
ð11Þ
and Z t��ðtÞ
t�h
_xT1 ðsÞR _x1ðsÞds
�1
h
Z t��ðtÞ
t�h
_xT1 ðsÞdsR
Z t��ðtÞ
t�h
_x1ðsÞds ð12Þ
then Equation (9) becomes
M � 2xT1 ðtÞP _xT1 ðtÞ þ �xT1 ðtÞPx1ðtÞ þ h2 _xT1 ðtÞR _x1ðtÞ
�
h�x1ðtÞ � x1ðt� �ðtÞÞ
�TR�x1ðtÞ � x1ðt� �ðtÞÞ
���x1ðt� �ðtÞÞ � x1ðt� hÞ
�T� R
�x1ðt� �ðtÞÞ � x1ðt� hÞ
�ie��h
þ xT1 ðtÞðEþ S Þx1ðtÞ � xT1 ðt� hÞEx1ðt� hÞe��h
� ð1� d ÞxT1 ðt� �ðtÞÞSx1ðt� �ðtÞÞe��h
� bwTðtÞwðtÞ: ð13Þ
Using the descriptor method as in Fridman (2001) and
the free-weighting matrices technique from He, Wang,
Lin, and Wu (2007)
0� 2ðxT1 ðtÞPT2 þ _xT1 ðtÞP
T3 Þ½� _x1ðtÞþ ðA11�A12KC1Þx1ðtÞ
þ ðAd11�Ad12KC1Þx1ðt� �ðtÞÞþB11wðtÞ�, ð14Þ
where matrix parameters P2,P3¼ "P22Rn�m are
added to the right-hand side of (13). Setting �ðtÞ ¼colfx1ðtÞ, _x1ðtÞ, x1ðt� hÞ, x1ðt� �ðtÞÞ,wðtÞg, then M�
�T(t)��(t)� 0 if the matrix �5 0. Multiplying
matrix � from the right and the left by diagfP�12 ,
P�12 ,P�12 ,P�12 , I g and its transpose, respectively, and
denoting
Q2 ¼ P�12 , bP ¼ QT2PQ2, bR ¼ QT
2RQ2,bE ¼ QT2EQ2, bS ¼ QT
2SQ2
it follows �5 0, b�5 0 where
b� ¼b�11 b�12 0 b�14 b�15 b�22 0 b�24 b�25 b�33 b�34 0
b�44 0
b�55
26666664
377777755 0, ð15Þ
and
b�11 ¼ ðA11 � A12KC1ÞQ2 þ �bPþQT2 ðA11 � A12KC1Þ
T
þ bEþ bS� bRe��h,b�12 ¼ bP�Q2 þ �QT2 ðA11 � A12KC1Þ
T,b�14 ¼ ðAd11 � Ad12KC1ÞQ2 þ bRe��h,b�15 ¼ B11, b�22 ¼ �"Q2 � "QT2 þ h2bR,b�24 ¼ "ðAd11 � Ad12KC1ÞQ2, b�25 ¼ "B11,b�33 ¼ �ðbEþ bRÞe��h, b�34 ¼ bRe��h,b�44 ¼ �2e��hbR� ð1� d ÞbSe��h, b�55 ¼ �bI:
ð16Þ
4 X. Han et al.
Downloaded By: [Steklyar, Inna] At: 07:06 15 November 2010
Select the LMI variable Q2 in the following form
Q2 ¼Q11 Q12
Q22M �Q22
� �, ð17Þ
where Q22 is a ( p�m)� ( p�m) matrix, M is a
( p�m)� (n� p) tuning matrix and � is a tuning
parameter to be selected by the designer. It follows that
KC1Q2 ¼ KQ22M �KQ22
� �:
Defining Y¼KQ22 it follows that
KC1Q2 ¼ YM �Y� �
: ð18Þ
To construct K, substitute (18) into (16) to yieldb�11 ¼ A11Q2 � A12½Y �Y � þQT2A
T11 þ �
bP,� ½YM �Y �TAT
12 þbEþ bS� bRe��h,b�12 ¼ bP�Q2 þ "Q
T2A
T11 � "½YM �Y �TAT
12,b�14 ¼ Ad11Q2 � Ad12½YM �Y � þ bRe��h,b�15 ¼ B11, b�22 ¼ �"Q2 � "QT2 þ h2bR,b�24 ¼ "Ad11Q2 � "Ad12½YM �Y �, b�25 ¼ "B11,b�33 ¼ �ðbEþ bRÞe��h, b�34 ¼ bRe��h,b�44 ¼ �2e��hbR� ð1� d ÞbSe��h, b�55 ¼ �bI:
ð19Þ
The following proposition can now be stated:
Proposition 4.2: Given scalars h4 0, d5 1, �4 0,
", �, b and a matrix M2R( p�m)�(n�p), if there exist
(n�m)� (n�m) matrices bP4 0, bE � 0, bS � 0, bR � 0
and matrices Q222R( p�m)�( p�m), Q112R
(n�p)�(n�p),
Q122R(n�p)�( p�m), Y2Rm�( p�m) such that the LMI
(15) with matrix entries (19) holds, then the reduced-
order system (7), where K ¼ YQ�122 , is exponentially
attracted to the ellipsoid
xT1 ðtÞPx1ðtÞ �b
�D2, ð20Þ
where P ¼ Q�T2bPQ�12 , for all differentiable delays
0 � �ðtÞ � h, _�ðtÞ � d5 1. Moreover, the reduced
order dynamics (7) is exponentially stable for all
piecewise-continuous delays 0� �(t)� h, if the LMI
(15) is feasible with bS ¼ 0.
Remark 1: Since the LMI (15) is affine in the system
matrices A, Ad and B1, the results are applicable to the
case where these matrices are uncertain. Denote �¼
[A Ad B1] and assume that �2Co{�j, j¼ 1, . . . ,N},
namely � ¼PN
j¼1 fj ðtÞ�j for some 0 � fj ðtÞ � 1,PNj¼1 fj ðtÞ ¼ 1, where the N vertices of the polytope
are described by �j ¼�Að j Þ A
ð j Þd B
ð j Þ1
�. One has to
solve the LMIs simultaneously for all the N vertices,
applying the same decision matrices for all vertices.
Example 4.3: Consider the following simple system
which is in regular form, with polytopic uncertaintiesand unknown (bounded) perturbations (t) and f(t)
_xðtÞ ¼
�3 2 1
2 1þ sinðx3ðtÞÞ 1
1 1 x22ðtÞ þ 1
264375xðtÞ
þ
0:5 0 0
0 1 0:2
�0:2 �0:5 1
264375xðt� �Þ
þ
0
0
1
264375 �uðtÞ þ
ðtÞx1ðtÞ þ 0:5wðtÞ
�0:5ðtÞx1ðt� �Þ � 0:5wðtÞ
0:2ðtÞx2ðtÞ þ wðtÞ
264375,
yðtÞ ¼0 1 0
0 0 1
� �xðtÞ ð21Þ
with 0�(t)� 2 and disturbance w(t)2 [�1, 1]. Thedelay is assumed to be time-varying. In order to
present (21) in the form of (1) with uncertain matrices,define the control variable �u(t) as follows:
�uðtÞ ¼ uðtÞ þ ðx22ðtÞ þ 1Þx3ðtÞ, ð22Þ
where u(t) is the sliding-mode control variable of theform (26). This change is possible because x2(t) andx3(t) are measured. Considering next sin(x3(t)) asuncertainty (t)¼ sin(x3(t))2 [�1, 1], the above systemis represented as a polytopic system with four verticesdefined by ¼1, ¼ 0 and ¼ 2
_xðtÞ ¼X4j¼1
fj ðtÞ Að j ÞxðtÞþA
ð j Þd xðt� �Þ
h iþB �uðtÞþB1wðtÞ,
ð23Þ
where
Að1Þ ¼
�3 2 1
2 2 1
1 1 0
264375, Að2Þ ¼
�3 2 1
2 0 1
1 1 0
264375,
Að3Þ ¼
�1 2 1
2 2 1
1 1:4 0
264375, Að4Þ ¼
�1 2 1
2 0 1
1 1:4 0
264375,
Að1Þd ¼ A
ð2Þd ¼
0:5 0 0
0 1 0:2
�0:2 �0:5 1
264375,
Að3Þd ¼ A
ð4Þd ¼
0:5 0 0
�1 1 0:2
�0:2 �0:5 1
264375,
B ¼
0
0
1
264375, B1 ¼
0:5
�0:5
1
264375: ð24Þ
International Journal of Control 5
Downloaded By: [Steklyar, Inna] At: 07:06 15 November 2010
Note that state-feedback sliding-mode control of theabove system without unmatched disturbances andpolytopic uncertainties 0.2(t)x2(t) in _x3ðtÞ was con-sidered in Gouaisbaut et al. (2004). The work employsLMI methods for the solution of the existence prob-lem, and is suitable for uncertain systems where thepolytopic uncertainties appear only in the subsystem(7). The control law is derived based on the assumptionthat those states varying in the span of the controlinput are bounded so that a large enough switchinggain can induce the sliding motion in finite time. Anappropriate switching gain must usually be determinedby trial and error.
The advantage of the proposed method is that itfacilitates analysis of polytopic uncertainties appearingin the input channel and the switching gain is derivedfrom LMIs, ensuring finite time reachability onto thesliding surface with a prescribed decay rate.
The initial function is taken as x(t)¼ [1, 1�1]T fort2 [��, 0]. To construct K for the reduced-order system(7) according to Proposition 4.2, the parameter settingsin the LMI (15) with entries (19) are selected with thedelay upper bound h¼ 1 s and the rate of change ofthe time-varying delay _� � d ¼ 0:1. For �¼ 2, "¼ 0.3,M¼ 2 and choosing �¼ 0.1, b¼ 0.005, then it isobtained that the LMI variables
bP ¼ 949:4 39:5
925:9
" #, Y ¼ 1237:4,
Q22 ¼ 175, K ¼ 7:07:
Once a stable sliding-mode dynamics has beendesigned, the next step is to find a controller whichensures the closed-loop system reaches the prescribedsliding surface in finite time. This will now beconsidered in general terms.
5. Reachability problem
It can be shown (Edwards et al. 2001) that thefollowing system transformation and control structure
exist such that z(t)¼T1xr(t), where T1 ¼In�m 0KC1 Im
h iso
that the system ð �A, �Ad, �B,F �C Þ has the property
�A ¼�A11
�A12
�A21�A22
" #, �Ad ¼
�Ad11�Ad12
�Ad21�Ad22
" #,
�B ¼0
Im
" #, �B1 ¼
�B11
�B12
" #, F �C ¼ 0 Im
� �, ð25Þ
where z1(t)¼ x1(t), z2(t)¼ s(t), �A11¼A11�A12KC1 and�Ad11¼Ad11�Ad12KC1 exhibit the reduced-ordersliding-mode dynamics. Also, �C ¼ ½0 �T �, where�T 2 Rp�p is nonsingular. The control law is
considered as
uðtÞ ¼ �GyðtÞ � vyðtÞ, ð26Þ
where
G ¼ G1 G2
� ��T�1 ð27Þ
vyðtÞ ¼�
FyðtÞ
kFyðtÞk, if FyðtÞ 6¼ 0,
0, otherwise,
8<: ð28Þ
where G12Rm�( p�m), G22R
m�m, F¼ [K Im]T�1. The
uncertain system (1) becomes
_zðtÞ ¼ �AzðtÞ þ �Adzðt� �ðtÞÞ þ �BuðtÞ þ �B1wðtÞ: ð29Þ
Closing the loop in the system (29) with the control law
(26) yields
_zðtÞ ¼ A0zðtÞ þ �Adzðt� �ðtÞÞ � �BvyðtÞ þ �B1wðtÞ, ð30Þ
where A0 ¼ �A� �BG �C. Let �P be a symmetric positive
definite matrix partitioned conformably with (25)
so that �P ¼�P1 00 �P2
h i. It follows that �P �B ¼ ðF �C ÞTP2 and
from (25) Fy(t)¼ z2(t). It can be shown that
¼ �PA0 þ AT0
�P
¼
�P1�A11 þ �AT
11�P1
�P1�A12 þ ð �A21 � G1C1Þ
T �P2
�P2
�A22 þ �AT22
�P2
� �P2G2 � ð �P2G2ÞT
( )26643775
¼�P1
�A11 þ �AT11
�P1�P1
�A12 þ �AT21
�P2 � ðL1C1ÞT
�P2�A22 þ �AT
22�P2 � L2 � ðL2Þ
T
" #,
ð31Þ
where L1 ¼ �P2G1 and L2 ¼ �P2G2. A stability condition
for the full order closed-loop system can be derived
using the following Lyapunov–Krasovskii functional
VðtÞ ¼ zTðtÞ �PzðtÞ þ
Z t
t�h
e ��ðs�tÞzTðsÞ �EzðsÞds
þ
Z t
t��ðtÞ
e ��ðs�tÞzTðsÞ �SzðsÞds
þ h
Z 0
�h
Z t
tþ�
e ��ðs�tÞ _zTðsÞ �R _zðsÞds d�, ð32Þ
where �E� 0, �S � 0 and �R ¼�R1 00 0
h iwhere �R1 � 0 (as it is
desired to determine a stability condition for the time-
delay system which is delay independent of z2(t)). Then
�M ¼ _Vþ ��V� �bwTðtÞwðtÞ
� 2zTðtÞ �P _zTðtÞ þ ��zTðtÞ �PzðtÞ þ h2 _zTðtÞ �R _zðtÞ
�
hðzðtÞ � zðt� �ðtÞÞÞT �RðzðtÞ � zðt� �ðtÞÞÞ
6 X. Han et al.
Downloaded By: [Steklyar, Inna] At: 07:06 15 November 2010
þ ðzðt� �ðtÞÞ � zðt� hÞÞT
� �Rðzðt� �ðtÞÞ � zðt� hÞÞie� ��h
þ zTðtÞð �Eþ �S ÞzðtÞ � zTðt� hÞ �Ezðt� hÞe� ��h
� ð1� d ÞzTðt� �ðtÞÞ �Szðt� �ðtÞÞe� ���ðtÞ � �bwTðtÞwðtÞ:
ð33Þ
Substitute the right-hand side of Equation (30) into (33).
Setting &(t)¼ col{z(t), z(t� h), z(t� �(t)),w(t)}, then
_VðtÞ � &ðtÞT�h&ðtÞ þ h2 _zTðtÞ �R _zðtÞ
þ 2zT �P �Bð �B12wðtÞ � vyðtÞÞ5 0 ð34Þ
is satisfied if &TðtÞ�h&ðtÞ þ h2 _zTðtÞ �R _zðtÞ5 0 and
2zT �P �Bð �B12wðtÞ � vyðtÞÞ5 0, where
�h ¼
�11 0 �P �Adþ �Re� ��h�P1
�B11
0
" # �22 �Re� ��h 0
�2e� ��h �R�ð1� d Þ �Se� ��h 0
� �bI
266666664
377777775ð35Þ
with
�11 ¼ þ �� �Pþ �Sþ �E� �Re� ��h; �22 ¼ �ð �Eþ �RÞe� ��h:
Setting �(t)¼ col{z(t), z(t� h), z(t� �(t)),w(t), vy(t)}
and �I ¼ Iðn�mÞ 0� �T
, it is obtained that
h2 _zTðtÞ �R _zðtÞ
¼ zTðtÞAT0 þ zTðt� �ðtÞÞ �AT
d � vTy ðtÞ�BT þ wTðtÞ �BT
1
h i� h2 �R A0zðtÞ þ �Adzðt� �ðtÞÞ � �BvyðtÞ þ �B1wðtÞ
� �
¼ �TðtÞ
AT0
0
�ATd
�BT1
�BT
26666664
37777775 �Ih2 �R1�IT
AT0
0
�ATd
�BT1
�BT
26666664
37777775
T
�ðtÞ: ð36Þ
Using the Schur complement, �TðtÞ�h�ðtÞ þh2 _zTðtÞ �R _zðtÞ5 0 holds if
�h
hAT0
Iðn�mÞ
0
� ��R1
0
h �ATd
Iðn�mÞ
0
� ��R1
h �BT1
Iðn�mÞ
0
� ��R1
� �R1
26666666666664
377777777777755 0 ð37Þ
for some ��4 0, �b4 0 and 0� �(t)� h, i.e. to ensure the
exponential attractiveness of (30) to the ellipsoid
zTðtÞ �PzðtÞ ��b��D
2. Given the control structure in (27),
then
2zTðtÞ �P �Bð �B12wðtÞ � vyðtÞÞ
¼ 2zT2 ðtÞ�P2ð �B12wðtÞ � vyðtÞÞ
� �2� �P2kz2ðtÞk þ 2 �P2k �B12kkz2ðtÞkD
5 0:
The latter inequality implies exponential attractivity
of the ellipsoid zTðtÞ �PzðtÞ ��b��D
2, thus for t!1,
zTðt� �ðtÞÞ �Pzðt� �ðtÞÞ ��b��D
2 holds. The following
proposition can now be stated:
Proposition 5.1: Given scalars h4 0, d5 1, ��4 0,�b4 0, assume there exist n� n matrices �P ¼ diagf �P1,�P2g4 0 with P22R
m�m, �E � 0, �S � 0, a (n�m)�
(n�m)-matrix �R1 � 0, L12Rm�( p�m), L22R
m�m such
that LMI (37) is feasible. Then for �4 k �B12kD the
closed-loop system (30), where G1 ¼ �P�12 L1, G2 ¼�P�12 L2, is exponentially attracted to the ellipsoid
zTðtÞ �PzðtÞ ��b��D
2 for all �(t)2 [0, h]. Consequently it
also holds that zTðt� �ðtÞÞ �Pzðt� �ðtÞÞ ��b��D
2 for t!1.
Denote
AL0 ¼ ½0 Im�A0, �AL
d ¼ ½0 Im� �Ad, ¼�b
��D2: ð38Þ
Given �14 0, �24 0, conditions will now be derived
that guarantee the solutions of (25) satisfy the bound
kAL0 zðtÞk5 �1, kA
Ld zðt� �ðtÞÞk5 �2 ð39Þ
for t!1. The following inequalities
zTðtÞðAL0 Þ
TðAL
0 ÞzðtÞ
� �21zTðtÞ �PzðtÞ
zTðt� �ðtÞÞðALd Þ
TðAL
d Þzðt� �ðtÞÞ
� �22zTðt� �ðtÞÞ �Pzðt� �ðtÞÞ
ð40Þ
guarantee (39). Hence equivalently
ðAL0 Þ
TðAL
0 Þ ��21
�P
, ðAL
d ÞTðAL
d Þ ��22
�P
ð41Þ
or by Schur complements
��21
�P
ðAL
0 ÞT
�I
24 355 0,��22
�P
ðAL
d ÞT
�I
24 355 0:
ð42Þ
International Journal of Control 7
Downloaded By: [Steklyar, Inna] At: 07:06 15 November 2010
Theorem 5.2: Given scalars ��4 0, �b4 0, let thereexist n� n matrices �P ¼ diagf �P1, �P2g4 0, �E � 0,�S � 0, a (n�m)� (n�m)-matrix �R1 � 0, L12
Rm�( p�m), L22R
m�m such that the LMI (37) is feasiblefor 0� �(t)� h, _�ðtÞ � d5 1. Let �1 and �2 satisfy (42)with the notation given in (38). Then for
�4 kB12kDþ �1 þ �2 ð43Þ
an ideal sliding motion takes place on the surface S. Theclosed-loop system (26), (29) is ultimately bounded by
lim supt!1 zTðtÞ �PzðtÞ ��b
��D2:
Proof: Substituting the control law it follows from(29) that
_sðtÞ ¼ F �CA0zðtÞ þ F �C �Adzððt� �ðtÞÞ þ ð �B12wðtÞ � vyðtÞÞ:
Let Vc :Rm!R be defined by VcðsÞ ¼ sTðtÞ �P2sðtÞ. It
follows that
�P2F �CA0 ¼ �P2AL0 ,
�P2F �CAd ¼ �P2ALd :
Starting from initial condition z(t0), it can be verifiedthat there exists t14 0 such that for all t� t1,
_VcðsÞ ¼ 2sTðtÞ �P2AL0 zðtÞ þ 2sTðtÞ �P2A
Ld zðt� �ðtÞÞ
þ 2sTðtÞ �P2ð �B12wðtÞ � vyðtÞÞ
� 2ksðtÞkk �P2kðkAL0 zðtÞk þ kA
Ld zðt� �ðtÞÞkÞ
� 2ð�1 þ �2ÞksðtÞkk �P2k
5 �2�ksðtÞk, ð44Þ
where � ¼ �1 þ �2 � kAL0 zðtÞk � kA
Ld zðt� �ðtÞÞk. A slid-
ing motion will thus be attained in finite time. œ
Remark 2: Since LMIs (15), (37) and (42) are affine inthe system matrices A, Ad and B1, the results are
applicable to the case where these matrices are uncer-
tain with polytopic type uncertainties (see Remark 1).
One has to solve the LMIs simultaneously for all the N
vertices, applying the same decision matrices for all
vertices. In contrast to the existing methods in the
literature (Gouaisbaut et al. 2004; Seuret, Edwards,
Spurgeon, and Fridman 2009), polytopic type uncer-
tainties can be incorporated in all the blocks ofA,Ad,B1
and not only in A11, Ad11 because the switching gain �(and not only the sliding surface) is found using LMIs.
Example 5.3: Following on from Example 4.3, where
a sliding surface prescribing stable dynamics has been
designed for the uncertain system (21), then the control
law in (27) will have the sliding function matrix
F¼ [7.07, 1]. A control gain G must be designed
which will bring the closed-loop system into a bounded
region centred at the sliding surface. Setting �� ¼ 0:3,�b ¼ 5 in Proposition 5.1, it is obtained that
�P¼
22:4 �12:8 0
29 0
0:68
26643775, L1 ¼�6:08, L2 ¼ 85:4,
which gives
G ¼ �9, 126:6� �
�T�1, where �T ¼1 0
�7:07 1
� �:
Once the state of the closed-loop system has entered
the sliding patch zTðtÞ �PzðtÞ ��b��D
2, the switching gain
�¼ 753 derived from LMI (42) will ensure the sliding
surface is reached in finite time. Figure 1 shows that
the sliding surface is reached in finite time and the
outputs of the system are stable with ultimate bound
ky(t)k� 0.2.Sliding-mode control has the advantage over
linear control for its absolute rejection of the
0 1 2 3 4 5 6 7 8 9 10−7
−6
−5
−4
−3
−2
−1
0
1
2
Time (s)
Out
puts
4 6 8 10−0.4
−0.2
0
0.2
0.4
0 2 4 6 8 10−5
0
5
10
15
20
25(b)(a)
Time
Slid
ing
surf
ace
Figure 1. Closed-loop response with delay h¼ 1 s: (a) outputs and (b) sliding surface.
8 X. Han et al.
Downloaded By: [Steklyar, Inna] At: 07:06 15 November 2010
matched uncertainties. To verify the statement, sup-
pose there is a change of the matched disturbance at
time 10 s of magnitude from 1! 50, then the sliding
surface remains unaffected so that B1¼ [0 0 50]T. While
keeping the same control parameters obtained so far
comparisons between using sliding-mode control and
only the linear control G for the new closed-loop
design with only matched disturbances are made in
Figure 3. As can be seen, the sliding-mode control is
more robust than the linear control to matcheddisturbances. The difference between using the linearpart of the control G alone and SMC with switchingfor a system with unmatched disturbances can bedemonstrated below. For the same uncertain system,suppose the unmatched disturbances are changed fromB1¼ [0.5,�0.5, 1]T to B1¼ [2,�2, 1]T after the initial10 s. Using the linear control G¼ [25, 4] alone in thefeedback, the responses for the outputs y(t), slidingfunction s(t) and control input u(t) are plotted inFigure 2(a). As can be seen the outputs ky(t)k� 1.4,sliding function is bounded as ks(t)k� 1.16 and controlinput ku(t)k� 4.7. If SMC is used with the same lineargain and a switching gain �¼ 7, the system responsesare plotted in Figure 2(b), where ky(t)k� 0.81, s(t)¼ 0,ku(t)k� 7. Therefore for a system with unmatcheddisturbances, SMC can give an ideal sliding surfaceresponse rather than the bounded sliding functiongiven by its linear control part. As a result, a smallerbound of the outputs can be obtained. For linearcontrol to yield the similar bound on the outputs as bySMC, the linear gain needs to be increased fromG¼ [25, 4] to G¼ [134, 21] as seen in Figure 2(c), whereky(t)k� 0.83. To conclude, for a linear control to givesimilar output performance in the presence ofunmatched disturbances, the linear gain needs to belarger, but not substantially larger than the linear partof the SMC. In another words, a linear control designfor a system with unmatched disturbances can give asimilar output bound as SMC if the linear control islarge enough.
6. Autonomous vehicle control
Autonomous vehicles are expected to operate effec-tively in time-varying and uncertain conditions. Here acase study described in Yao, Spurgeon, and Edwards(2006) is considered, where the elevation angle of thegun barrel of a vehicle in space should be maintainedwhile the hull of the vehicle is subject to externaldisturbances resulting from the motion of the vehicleacross rough terrain. To meet the high controlspecifications, sliding-mode control has been consid-ered within the application domain for its robustnessagainst friction and disturbances and its ease ofimplementation for motor drive control. The existenceproblem must determine a sliding surface that mini-mises the ultimate bound of the reduced-order dynam-ics in the presence of time-varying state delay andunmatched disturbances relating to frictional effects.A fully nonlinear simulation model of the system isavailable for controller analysis and testing (Yao et al.2006). The model is physically based and is known to
0 2 4 6 8 10 12 14 16 18 20−2
0
2(a)
(b)
(c)
X : 17.5Y : 1.393
Out
puts
0 2 4 6 8 10 12 14 16 18 20−2
0
2
X : 18.08Y : 1.16
Slid
ing
func
tion
0 2 4 6 8 10 12 14 16 18 20
−5
0
5
X : 18.17Y : −4.683
Inpu
t
0 2 4 6 8 10 12 14 16 18 20−1
0
1X : 17.91Y : 0.8051
Out
puts
0 2 4 6 8 10 12 14 16 18 20−10
0
10
Slid
ing
surf
ace
0 2 4 6 8 10 12 14 16 18 20−50
0
50
X : 17.42Y : 7
Con
trol
inpu
t
0 2 4 6 8 10 12 14 16 18 20−1
0
1X : 17.7Y : 0.8334
Out
puts
0 2 4 6 8 10 12 14 16 18 20−0.5
0
0.5
X : 16.74Y : −0.1176
Slid
ing
func
tion
0 2 4 6 8 10 12 14 16 18 20−5
0
5
X : 16.66Y : 2.642
Time
Inpu
t
Figure 2. Closed-loop response with linear control andSMC: (a) linear control with G, (b) SMC with linear part Gand (c) linear control with G.
International Journal of Control 9
Downloaded By: [Steklyar, Inna] At: 07:06 15 November 2010
represent with high fidelity the dynamics and behav-
iour of the real system:
_xðtÞ ¼
�Dm
JmN2�
Km
JmN
Dm
JmN0 0
1
N0 �1 0 0
Dm
J1N
Km
J1
�Dm�D12
J1
�K12
J1
D12
J10 0 1 0 �1
0 0D12
J2
K12
J2
�D12
J2
26666666666664
37777777777775|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
A
xðtÞ
þ
Kt
Jm0
0
0
0
266666664
377777775|fflfflffl{zfflfflffl}
B
uðtÞ
þ
�DmðN� 1Þ
JmN2�
1
Jm0
1
JmN20
N� 1
N0 0 0 0
DmðN� 1Þ
J1N0 �
1
J10
1
J10 0 0 0 0
0 0 0 0 0
2666666666664
3777777777775|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
B11
wðtÞ,
ð45Þ
where the state vector xðtÞ ¼ _�mðtÞ, �mbðtÞ, _�bðtÞ,�
�blðtÞ, _�l ðtÞ�T, and wðtÞ ¼ _�pðtÞ,!1mðtÞ,!1lðtÞ,
�Fmb�
signð _�mðtÞ � _�pðtÞÞ, Fmb signð _�bðtÞ � _�pðtÞÞ�T:The friction
related signals are
FmbðtÞ ¼ K�Dm
N_�mðtÞ þ
DmðN� 1Þ
N_�pðtÞ þ Km�mbðtÞ
,!1mðtÞ ¼ fms � signð�amðtÞ � Jm €�pðtÞÞ,
!1lðtÞ ¼ fls � signð�a1ðtÞ � J1 €�pðtÞÞ,
where the second and third states of the disturbance
from w(t) are a function of the friction level fd, where
fd can take values 0, 1, 2, 3 (Yao et al. 2006). Note
�mb ¼ 1N �m � �b þ ð1�
1N�p;
_�b breech velocity;�m motor position;_�l muzzle velocity;_�p pitch rate disturbance;Jm motor inertia;N gearbox of ratio;J1 elevation inertia on load one;
Km and Dm stiffness and damping
between the motor and the
load;K12 and D12 stiffness and damping
between the load one and
the load two;�am and �al applied torque to the motor
and the load;!1m and !1l motor friction and the load
friction;
0 1 2 3 4 5 6 7 8 9 10−7
−6
−5
−4
−3
−2
−1
0
1
2
Time (s)
Out
puts
2 4 6 8 10−0.2
−0.1
0
0.1
0.2
0.3
Sliding-mode control (y1)
Sliding-mode control (y2)
Linear control (y1)
Linear control (y2)
Figure 3. Comparison between sliding mode control and linear control in the presence of matched disturbance and delay.Available in colour online.
10 X. Han et al.
Downloaded By: [Steklyar, Inna] At: 07:06 15 November 2010
fms and fls motor coulomb friction andthe load coulomb friction;
u control input defined as thevoltage input to the poweramplifier;
_�p disturbance input defined aspitch rate disturbance;
!1m and signð _�m � _�pÞ motor friction disturbances;
!1l and signð _�b � _�pÞ load friction disturbances
The parameter values used in Yao et al. (2006)define
A¼
�338:14 �2:55� 107 50942 0 0
0:0066 0 �1 0 0
0:66 50000 �110:1 �15000 10
0 0 1 0 �1
0 0 7:69 11538 �7:69
26666664
37777775,
B¼
4523:1
0
0
0
0
26666664
37777775,
B11 ¼
�50604 �769 0 5:1 0
1 0 0 0 0
99 0 �0:01 0 0:01
0 0 0 0 0
0 0 0 0 0
26666664
37777775,
C¼
1 0 0 0 0
0 1 0 0 0
0 0 1 0 0
264375, ð46Þ
and the disturbance is known to be
k _�pk � 0:07, k!1mk � 0:5� fd, k!1lk � 10� fd,
kFmb signð _�m � _�pÞk � 4, kFmb signð _�b � _�pÞk � 4:
ð47Þ
The vehicle dynamics is augmented with an extra staterelated to the breech position, where the desired breechposition is zero. Denote Ca¼ [0 0 1 0 0], the state spacerepresentation of the augmented system is given by
Ac ¼0 �Ca
0 A
� �, Bc ¼
0
B
� �, Cc ¼
1 0
0 C
� �:
ð48Þ
6.1 Sliding-mode control design
A design which does not incorporate knowledge ofdelay effects is first performed to yield a benchmarklevel of performance. Firstly it is necessary to construct
K for the reduced-order system (7) according toProposition 4.2. The parameter settings in LMI (15)with entries (19) are selected as Ad¼ 0, h¼ 0 s. If
� ¼ 11:5, " ¼ 0:0000002, M ¼
0:15 0:015
0:21 0:06
0:009 0:0003
264375
and choosing �¼ 7.2, b¼ 0.0005, then it is obtainedthat
K ¼ 0:9, 28, 327� �
: ð49Þ
The poles of the corresponding reduced order systemare
�4:15 j106, �389:6 j160:7, �2823:7� �
: ð50Þ
The control law in (27) will have the slidingfunction matrix
F ¼ �327 0:0002 28 0:9� �
: ð51Þ
A control G is designed which will bring the closed-loop system into a bounded region centred about thesliding surface. Setting �� ¼ 0:8, �b ¼ 3:88� 10�6 inProposition 5.1, it is obtained that
G ¼ �1:02� 107, 7:6, 910610, 27834� �
: ð52Þ
The closed-loop poles of Ac�BcGCc are
�31257 �2810:4 �390:8 j160:9 �4:2 j106� �
:
The switching gain �¼ 1561, which is derived fromLMI (42), will ensure the sliding surface is reached infinite time. Figure 4 shows the position error and rateerror for fd¼ 1, 2, 3 using the proposed controller. Inthe original case study (Yao et al. 2006), an observerwas used to estimate the effect of the disturbance andthe equivalent control method was used to synthesisethe control law, which was augmented with an addi-tional PI control. With this strategy the position andrate errors were kep(t)k� 0.2� 10�3 rad and ker(t)k�0.01 rad/s respectively. Setting a fixed sampling fre-quency of 10 kHz and choosing ode3 solver insimulink, kep(t)k� 0.8� 10�5 rad, ker(t)k� 0.4�10�3 rad/s was achieved, as seen in Figure 4, forfd¼ 1, 2, 3 with the proposed control scheme. Theoutput feedback sliding-mode control approach pre-sented in this section has thus improved the trackingaccuracy over previous results in Yao et al. (2006).The ultimate bound of the outputs is a function of theunmatched disturbance, but it can be seen that theeffect of the friction disturbance on the controlperformance after changing fd¼ 1,! 3 is very small.
Speed control of a motor in the presence ofuncertainties such as friction normally exhibit delaysdue to the fact that the mechanical response of the
International Journal of Control 11
Downloaded By: [Steklyar, Inna] At: 07:06 15 November 2010
motor is slower than the electrical command. The sizeof the delay will depend upon the physical parametersof the actuator and can vary from milliseconds toseveral seconds, depending on the application. To takeaccount of such delay effects in the actual system forthe control design purpose, it is desirable to introducea system model incorporating delay into the systemmodel used for design. This will provide a means toanalyse the potential delay effect on the stability of theclosed-loop system at the design stage. Assume thedelay matrix
Ad ¼
�10 20 0 0 0
0:007 0 0:1 0 0
0 20 �2 1 0
0 0 0:1 0 0:1
0 0 2 1 0
26666664
37777775and the augmented matrix Adc ¼
0 0
0 Ad
� �:
ð53Þ
The open-loop tests on system (46) with the delaymatrix Ad in (53) shows that the vehicle system withstate delay h� 3ms yields a breech position error ofkep(t)k� 0.01 rad/s as expected from the known systemresponse. The augmented linear system with delay hasdynamics close to those of the original plant.
Designing a controller without considering explic-itly possible delay effects within the control designprocess can lead to deterioration of the system perfor-mance and sometimes even instability. Suppose there isa constant delay h¼ 3ms in the system where the valuesof F and G are taken as in (51) and (52) respectively,with the delay distribution matrix in (53). In this casethe position error will increase from kep(t)k� 0.8� 10�5
to kep(t)k� 1.4� 10�3 rad. The closed-loop systembecomes unstable for h� 4ms when delay effects arenot incorporated in the design process.
A controller will now be designed based on a modelincorporating delay effects. Firstly to construct Kfor the reduced-order system (7) according toProposition 4.2, the parameter settings in LMI (15)with entries (19) are selected with the delay upperbound h¼ 10ms and the rate of change of the time-varying delay _� � d ¼ 0. If for
� ¼ 3, " ¼ 0:0013, M ¼
0:0018 0:002
0:22 0:22
7:54 1:1
264375
and choosing �¼ 0.9, b¼ 0.0048, then it is obtainedthat
K ¼ 0:01, 16, 16:8� �
: ð54Þ
The poles of the reduced-order system are
�5:23 j92:9, �71:3 j254:8, �477:8� �
: ð55Þ
Thus the control law in (27) will have the slidingfunction matrix
F ¼ �16:8, 0:0002, 16, 0:01� �
:
The control G is designed to bring the closed-loopsystem into a bounded region centred about thesliding surface. Setting �� ¼ 0:8, �b ¼ 3:88� 10�6 inProposition 5.1, it is obtained that
G ¼ �2:02� 106, 26:7, 1:94, 0:002� �
:
The closed-loop poles of Ac�BcGCc are
�120700, �481:1, �75:2 j253:2, �5:43 j92:9� �
:
The switching gain �¼ 68,632, which is derived fromLMI (42), will ensure the sliding surface is reachedin finite time. The initial function was chosen asx(t0� �)¼ 0 for � 2 [0 h] in the simulation. The closed-loop performance is shown in Figure 5 for fd¼ 1, 2, 3.The position and rate errors are kept within the bound
0 1 2 3 4 5−1
−0.8−0.6−0.4−0.2
00.20.40.60.8
1 x 10−5
Time (s)
Pos
ition
err
or (
rad)
fd=1fd=2fd=3
0 1 2 3 4 5−4
−3
−2
−1
0
1
2
3
4(b)(a)x 10−4
Time (s)
Rat
e er
ror
(rad
/s)
fd=1fd=2fd=3
Figure 4. Closed-loop response without delay: (a) position error and (b) rate error. Available in colour online.
12 X. Han et al.
Downloaded By: [Steklyar, Inna] At: 07:06 15 November 2010
kep(t)k� 0.8� 10�4 rad, ker(t)k� 0.1� 10�4 rad/s in thepresence of delay. Despite the effect of the frictiondisturbance on the nonlinear model which is not fullyrejected, the controller is seen to be robust to thedisturbance even in the presence of delay. This hasdemonstrated the efficiency of the proposed controlscheme on a system of practical interest.
7. Conclusion
The development of output feedback-based sliding-mode control schemes for systems in the presence ofstate delay and both matched and unmatched distur-bances has been presented. A descriptor Lyapunovfunctional approach has been used for switchingfunction design. The methodology has been imple-mented using LMIs and can give desirable sliding-mode dynamics. The advantage of the method is thatfor the first time and despite only output feedbackbeing available, not only the switching function isderived from LMIs but also the switching gainrequired to solve the reachability problem is deter-mined using LMIs. The method allows polytopicuncertainties to be included in all blocks of A, Ad, B1
and not only in A11, Ad11 as with other methods. Thisis novel even for systems without delay. As well as anexample incorporating polytopic uncertainties, themethodology has also been applied to a nonlinearautonomous vehicle control problem. Nonlinear sim-ulations show that the gun barrel is maintained at thedesired position, despite variation in the vehicle motioncaused by friction.
Acknowledgements
The authors gratefully acknowledge EPSRC support viaresearch grant EP/E 02076311.
References
Cao, Y.Y., Lam, J., and Sun, Y.X. (1998), ‘Static Output
Feedback Stabilization: A LMI Approach’, Automatica,
34, 1641–1645.Choi, H.H. (2002), ‘Variable Structure Output Feedback
Control for a Class of Uncertain Dynamic Systems’,
Automatica, 38, 335–341.Crusius, C.A.R., and Trofino, A. (1999), ‘Sufficient LMI
Conditions for Output Feedback Control Problems’, IEEE
Transactions on Automatic Control, 44, 1053–1057.Edwards, C. (2004), ‘A Practical Method for the Design of
Sliding Mode Controllers using Linear Matrix
Inequalities’, Automatica, 40, 1761–1769.Edwards, C., Akoachere, A., and Spurgeon, S.K. (2001),
‘Sliding Mode Output Feedback Controller Design using
Linear Matrix Inequalities’, IEEE Transactions on
Automatic Control, 46, 115–119.Edwards, C., and Spurgeon, S.K. (1995), ‘Sliding Mode
Stabilization of Uncertain Systems using only Output
Information’, International Journal of Control, 62,
1129–1144.
El-Khazali, R., and Decarlo, R.A. (1995), ‘Output Feedback
Variable Structure Controllers’, IEEE Transactions on
Automatic Control, 31, 805–816.
Fernando, C., and Fridman, L. (2006), ‘Analysis and Design
of Integral Sliding Manifolds for Systems with Unmatched
Perturbations’, IEEE Transactions on Automatic Control,
51, 853–858.Fridman, E. (2001), ‘New Lyapunov–Krasovskii Functionals
for Stability of Linear Retarded and Neutral Type
Systems’, Systems and Control Letters, 42, 233–240.Fridman, E., and Dambrine, M. (2009), ‘Control Under
Quantization, Saturation and Delay: A LMI Approach’,
Automatica, 10, 2258–2264.Gouaisbaut, F., Blanco, Y., and Richard, J.P. (2004),
‘Robust Sliding Mode Control of Non-linear Systems
with Delay: A Design via Polytopic Formulation’,
International Journal of Control, 77, 206–215.Gouaisbaut, F., Dambrine, M., and Richard, J.P. (2002),
‘Robust Control of Delay Systems: A Sliding Mode
0 1 2 3 4 5−1
0
1x 10−4
Time (s)
Pos
ition
err
or (
rad)
fd=1fd=2fd=3
0 1 2 3 4 5–3
−2
−1
0
1
2
3(b)(a) x 10−3
Time (s)
Rat
e er
ror
(rad
/s)
fd=1fd=2fd=3
Figure 5. Closed-loop response with constant delay h¼ 10ms present in the plant where the control has been designed based on atime-delay system model: (a) position error and (b) rate error. Available in colour online.
International Journal of Control 13
Downloaded By: [Steklyar, Inna] At: 07:06 15 November 2010
Control Design via LMI’, Systems and Control Letters, 46,219–230.
Han, X., Fridman, E., Spurgeon, S.K., and Edwards, C.(2009), ‘On the Design of Sliding Mode Static OutputFeedback Controllers for Systems with State Delay’, IEEETransactions on Industrial Electronics, 56, 3656–3664.
He, Y., Wang, Q.G., Lin, C., and Wu, M. (2007), ‘DelayRange Dependent Stability for Systems with Time-varyingDelay’, Automatica, 43, 371–376.
Huang, D., and Nguang, S.K. (2006), ‘Robust H1 StaticOutput Feedback Control of Fuzzy Systems: A LMIApproach’, IEEE Transactions on Systems, 36, 216–222.
Jafarov, E.M. (2005), ‘Robust Sliding Mode ControllerDesign Techniques for Stabilisation of MultivariableTime-delay Systems with Parameter Perturbations andExternal Disturbances’, International Journal of System
Science, 36, 433–444.
Li, X., and DeCarlo, R.A. (2003), ‘Robust Sliding ModeControl of Uncertain Time Delay Systems’, International
Journal of Control, 76, 1296–1305.Seuret, A., Edwards, C., Spurgeon, S., and Fridman, E.(2009), ‘Static Output Feedback Sliding Mode ControlDesign via an Artificial Stabilizing Delay’, IEEE
Transactions on Automatic Control, 54, 256–265.Shaked, U. (2003), ‘An LPD Approach to Robust H2 andH1 Static Output Feedback Design’, IEEE Transactions
on Automatic Control, 48, 866–872.Yao, L., Spurgeon, S.K., and Edwards, C. (2006), ‘AnInvestigation of Novel Reconfiguration Methods for Use
in Autonomous and Semi Autonomous Vehicles’, in SEASDTC Technical Conference, Edinburgh.
Zak, S.H., and Hui, S. (1993), ‘On Variable Structure OutputFeedback Controllers for Uncertain Dynamic Systems’,
IEEE Transactions on Automatic Control, 38, 1509–1512.
14 X. Han et al.
Downloaded By: [Steklyar, Inna] At: 07:06 15 November 2010